Abstract: Various geometric vision problems can be solved optimally by minimizing L_∞-norm error.However，the approach based on L_∞-norm error is often sensitive to outliers.Although some improved algorithms can overcome the impact of outliers，it is too time-consuming.This paper proposes a new fast and precise method based on refining L_∞-norm minimization framework for the triangulation problem from images sequence.Experimental results have shown that the proposed method can achieve global optimization and is faster than traditional methods.

Key words: computer vision, triangulation, L_∞-minimization, outliers, Second-Order Cone Programming（SOCP）

摘要： 多视几何中的多种问题可以通过最小化L_∞范数误差获得全局最优解。但最小化L_∞范数误差算法的缺点是对外点敏感，相关的改进算法虽然可以克服外点带来的影响，但计算速度较慢。提出一种改进的最小化L_∞范数误差算法，用于从包含外点的图像序列中快速精确重建三维空间点。真实测试图像的实验结果证明该算法可以在包含外点的情况下获得空间点的全局最优解，相比其他算法速度有较大的提高。

关键词: 计算机视觉, 三维重建, 无穷范数最小, 外点, 二次锥面规划